Estimating the likelihood of epilepsy from clinically noncontributory electroencephalograms using computational...

Area of Science:

• Neuroscience
• Computational Biology
• Medical Diagnostics

Background:

• Epilepsy diagnosis relies on clinical assessment and electroencephalography (EEG).
• Identifying seizure susceptibility biomarkers from EEG is crucial for early diagnosis and treatment.
• Existing diagnostic methods can face challenges with accuracy and timely results.

Purpose of the Study:

• To validate a set of candidate seizure susceptibility biomarkers derived from routinely collected EEG data.
• To assess the robustness of these biomarkers in a large, multi-site cohort including epilepsy and non-epilepsy conditions.
• To determine the impact of potential confounding variables on biomarker performance.

Main Methods:

• A retrospective case-control study involving 814 EEG recordings from 648 subjects across eight UK sites.
• Calculation of eight computational markers (spectral, network-based, model-based) from each EEG recording.
• Development of ensemble-based classifiers and assessment of confounding variables using regression methods.

Main Results:

• Balanced accuracy of 68% was achieved across the cohort with clinically noncontributory normal EEGs (Sensitivity=61%, Specificity=75%).
• The study demonstrated a negative predictive value of 79% and an area under the ROC curve of 0.72.
• No significant impact of confounding variables such as age, gender, or treatment status on overall biomarker performance was found.

Conclusions:

• The validated set of EEG biomarkers can enhance clinical decision-making for seizure susceptibility.
• These biomarkers form a foundation for a decision support tool to reduce diagnostic delay and misdiagnosis rates.
• Future prospective studies are recommended to evaluate the diagnostic yield and time to diagnosis when using these biomarkers.